Measurement 132 (2019) 263–271

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Uncertainty analysis applied to electrical components diagnosis by

infrared thermography
R.A.M. Ferreira a, B.P.A. Silva a, G.G.D. Teixeira a, R.M. Andrade a, M.P. Porto a,b
a
Laboratório de Termometria do Departamento de Engenharia Mecânica da UFMG, Programa de Pós Graduação em Engenharia Mecânica, Universidade Federal de Minas Gerais,
Belo Horizonte, MG, Brazil
b
Bolsista do CNPq (303861/2017-7), Brazil

a r t i c l e i n f o a b s t r a c t

Article history: This work presents a discussion on the use of infrared thermography to detect degradation of electrical
Received 31 July 2018 components, highlighting the importance of uncertainty analysis into decision making process. Some
Accepted 13 September 2018 technical standards recommend to use temperature differences between similar components as an integ-
Available online 21 September 2018
rity indicative (a comparative approach). Temperature measurement from infrared cameras is dependent
of target surface emissivity and environmental conditions, which impact the measurement accuracy and
Keywords: lead comparative approaches to inconclusive results. We make laboratory experiments under controlled
Infrared thermography
conditions to demonstrate how emissivity affects the temperature uncertainty. We also performed a
Measurement uncertainty
Monte Carlo Method
thermographic inspection using NETA MTS-2011 for an electrical substation connector and discuss the
GUM influence that temperature uncertainty has on recommended actions suggested by this standard.
Uncertainty was determined using the Guide to the Expression of Uncertainty in Measurement (GUM)
and the Monte Carlo Method (MCM). We concluded that uncertainty measurement analysis of thermo-
graphic inspections is a crucial step on preventive maintenance of electrical components.
! 2018 Elsevier Ltd. All rights reserved.

1. Introduction exactly these both alternatives and recommends predetermined

Thermography has a wide application in detecting degradation
of electrical components in substations. This is because electrical
deterioration is manifested in terms of temperature increase, and
thermal images are useful to identify hot spots. Thermography
has improved the efficiency on maintenance of electrical installa-
tions, allowing long distances inspections without physical contact
(safety) and not requiring any equipment shutdown (no produc-
tion loss).
But, temperature measured by itself is not a self indicative of
the components condition, since the environmental variations
(wind velocity, irradiance, humidity, etc.) have direct influence
on the thermodynamic equilibrium. In this context, some methods
propose to use temperature difference between similar objects in
the same scene as an integrity criterion. In these methods, the
selected objects are placed in the same thermal image frame,
where one component is the reference and the other is under eval-
uation. In other technical recommendations, temperature differ-
ence between the component and environment is used as
integrity criterion. Technical standard NETA MTS-2011 [1] suggests
actions for each given temperature difference interval, as can be
seen in Table 1.
In some situations, thermographers develop their own tables
based on their experience in a particular substation or component
looking for more reliable results. However, there is no scientific
guarantee of results obtained without an appropriate uncertainty
analysis [2].
Thermography technique has been applied for different
areas of engineering [3–7], and methods to estimate
uncertainty in thermal images are well established [8–11]. In
case of electrical components failure detection, studies on
thermography can also be found in the literature [12–14],
including smart monitoring systems [15,16]. However, none
of them presents an uncertainty analysis, nor highlight its
importance to diagnosis evaluation and in the decision making
process.
Looking to fill out this gap, in this paper we present a method-
ology based on GUM and Monte Carlo Method (MCM) to estimate
the temperature uncertainty of electrical components, highlighting
the influence of target surface emissivity on results.

https://doi.org/10.1016/j.measurement.2018.09.036
0263-2241/! 2018 Elsevier Ltd. All rights reserved.
264 R.A.M. Ferreira et al. / Measurement 132 (2019) 263–271

Nomenclature

c sensitivity coefficient U expanded uncertainty

d distance, m v eff degrees of freedom
dhigh absolute difference of upper limits of GUM and MCM xi input variable i estimated value
coverage intervals Xi input variable i true value
dlow absolute difference of lower limits of GUM and MCM y output variable estimated value
coverage intervals Y output variable true value
DTR digital temperature resolution y
b output variable average value
g ðxi Þ probability density function of variable xi
kp coverage factor Greek
M number of random values generated in MCM dt sensitivity for a non-uniform FOV, K
ME minimum error e emissivity
MSR measurement spatial resolution gi output variable random value
MU measurement uniformity ni input variable random value
n number of measurand observations s atmospheric transmissivity
N number of input variables in f model
NGE noise generated error Subscripts
r correlation coefficient amb ambient
R electrical resistance, lX
atm atmospheric
RE repeatability C1 degraded connector
s standard deviation C2 new connector
S detector signal, W m#2 intr intrinsic source
STRF slit temperature response function
max coverage interval upper limit
tp Student’s t-distribution factor min coverage interval lower limit
DtNETD thermal resolution, K ob object
D Tb average temperature difference, K p confidence level
T temperature, K
p1 ,p2 suspect point reading
TS temperature stability ref reference
DT temperature difference, K refl reflected
u standard uncertainty
uc combined standard uncertainty

Table 1
Criteria and recommended actions by NETA MTS-2011 [1].

DT DT amb Recommended action

1 "C–3 "C 1 "C–10 "C Possible deficiency; warrants investigation.
4 "C–15 "C 11 "C–20 "C Indicates probable deficiency; repair as time
permits.
– 21 "C–40 "C Monitor until corrective measures can be
accomplished.
>15 "C >40 "C Major discrepancy; replace immediately.

2. Temperature measurement by thermography

Fig. 1. Schematic view of temperature measurement with a thermal imaging

During a thermographic inspection, the radiation captured by
the imager lenses is comprised of three parts. The mathematical
relation to describe the radiation absorbed by infrared sensors (S)
is given by Eq. (1) [17]
S ¼ esSob þ sð1 # eÞSrefl þ ð1 # sÞSatm ;
where e is the object emissivity and s is the atmosphere transmis-
sivity between the target’s surface and detector. Sob ; Sref and Satm are
the detectors input signals corresponding to the object, reflected
and ambient temperatures. Fig. 1 depicts all signals that compound
a inspection scenario from an object with emissivity, e, and temper-
ature T ob located at a distance d of an infrared camera. Then, the
imager converts all radiation inputs into an estimated value of
object temperature.
Atmospheric transmissivity can be considered a function
dependent of the object distance, with no representative loss of
accuracy. So, we conclude that the object temperature can be gen-
erally represented as
! "
T ob ¼ f e; d; T refl ; T atm ; S :
camera [17].
3. Measurement uncertainty
ð1Þ
3.1. GUM method
Guide to the Expression of Uncertainty in Measurement (GUM)
[18] divides uncertainties into two categories, Type A and Type B.
Type A is obtained from standard deviation, s, which is esti-
mated from a series of measurand readings. Type B evaluation is
characterized by choosing the best probability density function
ð pdf Þ to represent the measurements [18]. Both uncertainty types
obey probability distributions, and can be evaluated by statistical
methods.
In case of an indirect measurement, the output true value Y is
determined using input true values X i into the mathematical
relation
ð2Þ Y ¼ f ðX 1 ; X 2 ; . . . ; X N Þ: ð3Þ
 R.A.M. Ferreira et al. / Measurement 132 (2019) 263–271 265

X i values are not practical to obtain due to the limited accuracy

of instruments, thus estimated values xi are used into Eq. 3 to esti-
mate the output y
y ¼ f ðx1 ; x2 ; . . . ; xN Þ; ð4Þ
where each xi is associated to a standard uncertainty
uðx1 Þ; uðx2 Þ; . . . ; uðxN Þ defined as Type A or Type B. The combined
standard uncertainty uc ð yÞ is estimated by the Uncertainty Propaga-
tion Law, Eq. (5) [19]. A schematic representation of GUM procedure
is depicted in Fig. 2.
N
X N#1 X
X N
! " ! " Fig. 3. Schematic diagram representing the Monte Carlo Method to estimate
u2c ð yÞ ¼ c2i u2 ðxi Þ þ ci cj uðxi Þu xj r xi ; xj : ð5Þ
measurement uncertainty [19].
i¼1 i¼1 j¼iþ1

After determining the combined standard uncertainty, the

M
expanded uncertainty, U p , can be calculated by the product 1X
y¼
b y; ð9Þ
U p ¼ kp uc ð yÞ where kp is a coverage factor. Expanded uncertainty M r¼1 r
is the range in which the measured value is expected to be, for a
confidence level p. Therefore, the measurement result is expressed and the standard uncertainty is
by
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u M
u 1 X ! "2
Y ¼ y & Up : ð6Þ uð yÞ ¼ t y #b y : ð10Þ
M # 1 r¼1 r
The coverage factor kp that achieves the level of confidence p is
! " To generate the pseudo-random numbers, we used the Linear
obtained by assuming the approximation kp ¼ t p v eff where
! " Congruential Method [23], which is able to generate a sequence
t p v eff is a factor derived from Student’s t-distribution with effec-
of numbers with uniform distribution within the range from 0 to
tive degrees of freedom v eff , calculated by the equation of Welch-
1. Basically, this is a recursive method that starts from an initial
Satterthwaite [19]
number X 0 , called seed, and all following numbers given by
u4 ð yÞ
v eff ¼ P c u4 ð yÞ ; ð7Þ X iþ1 ¼ ðaX i þ bÞ mod m;
N i ð11Þ
i¼1 vi Ri ¼ X i =m;
and the expanded uncertainty is where a is a constant, b an increment, and m the modulus of the
! " recursive method. The current normalized element in the generated
U p ¼ tp v eff uc ð yÞ: ð8Þ
random sequence is represented by Ri .

3.2. Monte Carlo Method 3.3. Validation

Monte Carlo Method (MCM) is a class of numerical approaches
that uses random numbers to support statistical analysis [20].
MCM is considered one of the simplest methods to solve uncer-
tainty problems in a probabilistic framework [21], and is an alter-
native when the mathematical model is complex or does not meet
GUM required application criteria [22].
As can be seen in Fig. 3, a probability density function g xi is
defined for each one of the N input variables that comprises the
mathematical model Y ¼ f ðX 1 ; X 2 ,. . ., X N Þ. Then, a large number M
of random values ni is generated for each input variable, obeying
the probability distributions previously assumed. M sets of random
inputs are simultaneously evaluated by the f model, producing M
results gi used to obtain statistical properties and estimate the
probability density function g Y of the output variable Y.
Numerical results of Y are used to determine the average value
! "
y , standard uncertainty u b
b y , and coverage interval ½ymin ; ymax (,
which is based on the desired confidence level p. The average value
y , for M results ordered by yr and r ¼ 1; . . . ; M, is given by
b Thermal imager are calibrated under a controlled room temper-
Fig. 2. Schematic diagram representing GUM method to estimate measurement
uncertainty [19].
The Joint Committee for Guides in Metrology (JCGM) [22] rec-
ommends to compare results from GUM and MCM through Eq. 12
! "
dlow ¼ j y # U p # ymin j;
! " ð12Þ
dhigh ¼ j y þ U p # ymax j:
Absolute differences between interval dlow and dhigh must not
exceed a predetermined tolerance, d. Values of d are based on a ref-
erence significant digit nd given by the expression of the combined
standard uncertainty. In this paper is assumed a value of d ¼ 0:5 K,
applicable in temperature measurements, according to the recom-
mendations of [22].
3.4. Sources of measurement uncertainty
Sources of measurement uncertainty are divided into intrinsic
(see Fig. 4), related to the infrared camera internal components,
and extrinsic, due to the environmental conditions and measurand.
ature, which means that the camera sensors are maintained under
the same temperature along all calibration process. On the other
hand, during the thermographic inspections the ambient tempera-
ture is usually not the same, and sensors operate out of the calibra-
tion condition, affecting the result [24]. Temperature Stability (TS)
estimates the influence of ambient temperature on the camera
sensors. The distribution used to represent this intrinsic uncer-
tainty is the uniform distribution, given by
$ % TS
uTS Tb ob ¼ pffiffiffiffiffiffi : ð13Þ
12
266 R.A.M. Ferreira et al. / Measurement 132 (2019) 263–271

Measurement Uniformity (MU) estimates the range for all mea-

surement results in different measurand positions within the field
of view (FOV) of the thermal imager [24], as observed in Fig. 5 ðbÞ.
This uncertainty is estimated by a thermal imager parameter
defined as sensitivity to a nonuniform field of view dt

dt ¼ Tb max # Tb min ; ð15Þ

where Tb max and T

b min are the maximum and minimum mean temper-
atures obtained in each area of the thermogram, respectively. The
best estimation of MU is given as an uniform distribution where
MU ¼ dt
$ % MU
uMU Tb ob ¼ pffiffiffiffiffiffi : ð16Þ
12
Noise Generated Error (NGE) is defined as the temperature stan-
dard deviation due to internal noise, and is estimated based on the
NETD (Noise Equivalent Temperature Difference), also known as
thermal resolution, which is obtained using a blackbody source
[25].
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u $ %2
pffiffiffi u X Y
2u XX DT ij # D Tb
DtNETD ¼ t ; ð17Þ
2 i¼1 j¼1
HV
Fig. 4. Intrinsic sources of uncertainty during a thermographic inspection:
Temperature Stability (TS), Repeatability (RE), Digital Temperature Resolution
where DT ij is the temperature difference between consecutive ther-
(DTR), Minimum Error (ME), Measurement Spatial Resolution (MSR), Measurement
Uniformity (MU), and Noise Generated Error (NGE). mograms at the same position ði; jÞ, and D Tb represents the average
among all temperature differences. H and V are the pixels number in
Measurement Spatial Resolution (MSR) is defined as the mini- horizontal and vertical directions, respectively.
mum angular dimension of the measurand that produces some In case of Dt NETD ¼ NGE
influence on the temperature measurement result [24]. MSR can $ %
uNGE Tb ob ¼ DtNETD : ð18Þ
be estimated using the procedure depicted in Fig. 5 ðaÞ, consisting
on the observation of a blackbody source through a vertical slit.
Minimum Error (ME) uncertainty is estimated using a cali-
MSR is proportional to the angular slit dimension (a ¼ l=R) for
brated blackbody source in a similar configuration as depicted in
which the Slit Temperature Response Function (STRF) is 0.99 [24]
Fig. 5 ðcÞ, and is defined as the dispersion of the measured value
T slit # T amb T ob in respect to the true value T actual [24]. The contribution of ME
STRF ¼ : ð14Þ is also given by an uniform distribution
maxðT slit # T amb Þ
$ % d
ME
The uncertainty due to the MSR is often neglected because the uME Tb ob ¼ pffiffiffiffiffiffi ; ð19Þ
angular size of the measurand is in most cases greater than the 12
angular slit dimension a.

Fig. 5. Methods to estimate intrinsic sources of measurement uncertainty using a blackbody source: a) MSR b) MU and c) NGE.
 R.A.M. Ferreira et al. / Measurement 132 (2019) 263–271 267

n
X
d 1 As part of laboratory tests, we built a test bench comprised of an
ME ¼ MEi ;
n ð20Þ AC current source, aluminum cables, connectors (used alternately
i¼1
two H type connectors, see Fig. 6), a scientific infrared camera, K-
MEi ¼ j Tb actual;i # Tb ob;i j: type thermocouples connected to a data acquisition system, a
Repeatability (RE) is determined from the mean experimental microohmmeter to verify electrical resistance of the samples and
$ % a meteorological data acquisition system to measure ambient con-
standard deviation sRE Tb ob , using a similar procedure as shown
ditions, as described in Table 2 (Fig. 7).
in Fig. 5(c), by taking n observations of the object temperature T ob The selected H-type connectors (namely C1 and C2) have initial
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi resistances RC1 ¼ 199 lX and RC2 ¼ 85 lX and emissivities of 0.13
u n $ %2
u 1 X and 0.12, respectively. The difference between both samples is that
sRE ðT ob Þ ¼ t T ob;k # Tb ob ; C1 is degraded, which can be confirmed by the high value of elec-
n # 1 k¼1
ð21Þ trical resistance, whereas C2 is new, and it was not used before the
$ % sRE ðT ob Þ
uRE Tb ob ¼ pffiffiffi :
n
Digital Temperature Resolution (DTR) is a source of uncertainty
that accounts for the influence of limited resolution of digital chan-
nels on the minimum temperature difference that can be observed.
DTR value can be estimated by [24]

DT span
DTR ¼ ; ð22Þ
2k
where k is the bits number of the analog-digital converter, and a) Connector C1
DT span is the nominal temperature range of the camera. DTR distri-
bution is assumed as uniform
$ % DTR
uDTR Tb ob ¼ pffiffiffiffiffiffi : ð23Þ
12
Extrinsic uncertainty sources are due to emissivity, distance,
reflected temperature and atmospheric temperature.
Emissivity can be estimated by basically two comparative
approaches: using a known emissivity coating or tape, or with a
calibrated thermocouple. The first technique consists on applying b) Connector C2
a high emissivity coating in a small portion of the surface. The high
emissivity region can be used to determine the temperature accu- Fig. 6. H-type electrical connectors used during the experiments.
rately. As long as the temperature is the same all over the body sur-
face, we can target at the non-coated region adjusting the
emissivity until the temperature be the same. The second approach Table 2
Instruments specification for the experimental test bench.
is quite similar and consists on adjusting the emissivity until the
temperature observed by a thermal imager be the same as the Instrument Commercial model
one measured by a thermocouple, assumed as reference. Applying AC current source SMC LET-1000-RD
one of the aforementioned techniques to several points of the tar- Infrared camera Flir SC660
get surface is possible to obtain statistical properties of emissivity, Data acquisition system Agilent 34970A
Microohmmter Instrum MICROHM-10
as the mean and standard deviation values, also to estimate the
Scientific ambient monitor TESTO 622
probability density function of this variable.
Variables d; T refl e T atm are measured for each one experiment
made. The reflected temperature is assumed as the same as the
ambient temperature, which was obtained from a meteorological
data acquisition system.

4. Methodology

4.1. Experimental procedure

Emissivity has a direct influence in a radiometric measurement.

So, variations in the surface condition may lead to temperature
readings completely different during a thermographic inspection
in electrical components. In addition, weather conditions and
measurand-dependent variables also need to be known in order
to obtain a reliable diagnostic for an identified thermal anomaly.
During an outdoor thermographic inspection is difficult to sep-
arate all uncertainty components, mainly due to weather variabil-
ity (wind, humidity and solar irradiation), then first we decide to
make steady-state experiments in laboratory to understand the Fig. 7. Experimental test bench built to make thermographic analysis of electrical
indoor effects first [26]. components.
268 R.A.M. Ferreira et al. / Measurement 132 (2019) 263–271

experiments we have made. To determine the emissivity experi-

mentally, connectors C1 and C2 surfaces were partially covered
with a commercial coating, with e ¼ 0:97.
To perform the experiments in laboratory, we take the follow-
ing steps:

1. install the selected connector, C1 or C2;

2. measure the electrical resistance of the connector;
3. attach the thermocouples on the sample;
4. start the data acquisition system;
5. adjust the electrical current as 150 A;
6. record the environmental conditions;
7. monitor the steady-state by the thermocouples readings;
8. use the infrared camera to register thermograms (in the coated
and non-coated areas);
9. adjust the electrical current for 50 A higher and return to step 6;
500 A is the maximum current evaluated.
We repeat the aforementioned procedure for both H-type connec-
tors. For all measurements, the camera was positioned 30 ) out of
normal direction to avoid the influence of reflected radiation from
the thermographer into the results [27].
4.2. Measurement uncertainties
In this section we present GUM and MCM assumptions to esti-
mate the combined standard uncertainty of the temperature
measurements.
4.2.1. GUM assumptions
Eq. (5) presents the uncertainty sources that comprise the
mathematical relation of the combined standard uncertainty.
Including the intrinsic and extrinsic uncertainties of a thermo-
graphic measurement and simplifying Eq. (5), we obtain
u2c ðT ob Þ ¼ c2e u2e þ c2d u2d þ c2T refl u2T refl þ c2T atm u2T atm þ u2intr :
In Eq. (24), all variables were considered non-correlated. The uncer-
tainty component uintr comes from the combined intrinsic uncer-
tainties discussed in Section 3.4, and is determined by Eq. (25)
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uintr ¼ u2NGE þ u2TS þ u2RE þ u2DTR þ u2ME þ u2MU þ u2MSR :
Fig. 8. Histogram of emissivity measurements, and the probability distribution
function associated.
used in the experiments and its respective pdf estimated by poly-
nomial interpolation.
To generate a sequence of random numbers that obey the pdf,
we limit the range of possible values that the emissivity can
assume along the numerical simulation between 0.15 and 0.45.
With the set of possible values and associated probabilities
given by g ðxÞ function, the next step is to perform successive tests
using the Linear Congruential Method (LCM) to include random
numbers at the list being generated. The advantage of this proce-
dure lies in the use of a random number generator (as LCM), that
essentially provides a rectangular distribution, to generate random
numbers that obey a generic pdf.
Fig. 9 shows a histogram of generated values, provided with the
aforementioned algorithm, compared to a histogram of experimen-
tal values. In this case, we observed a deviation less than 3%
between numerical and experimental results.
Table 4 shows all assumed distributions, in case of MCM. Only
variables those were presented on Eq. (2) received an associated
pdf, also the digital signal S, which involves all intrinsic
ð24Þ
uncertainties.
ð25Þ

Table 3 presents the probability distributions we assume for all

the uncertainties sources.

4.2.2. MCM assumptions

In this work, the probability density function of emissivity was
obtained experimentally by comparison using the contact temper-
ature method. Fig. 8 presents the histogram obtained from a set of
50 measurements along the surfaces of both H-type connectors

Table 3
Fig. 9. Comparison between experimental and simulated values of emissivity; the
Probability distribution functions assumed during GUM evaluation.
latter is provided from a routine developed to randomly generate emissivity values
Uncertainty Sources Distribution obeying the probability density function g ðxÞ.

Emissivity, e Uniform
Target Distance, d Uniform
Table 4
Reflected Temperature, T refl Uniform
Assumed probability distributions for the MCM.
Atmospheric Temperature, T atm Uniform
Noise Generated Error, NGE Uniform Uncertainty Sources Distribution
Temperature Stability, TS –
Measurement Repeatability, RE Normal Emissivity, e g ð xÞ
Digital Temperature Resolution, DTR Uniform Target Distance, d Uniform
Minimum Error, ME Uniform Reflected Temperature, T refl Uniform
Measurement Spatial Resolution, MSR – Atmospheric Temperature, T atm Uniform
Measurement Uniformity, MU Uniform Digital Signal, S Normal
 R.A.M. Ferreira et al. / Measurement 132 (2019) 263–271 269

&8:5 K, respectively, for 450 A electrical current value, as shown

in Fig. 11b.

5.2. Infrared thermographic inspection in an electrical substation

We also performed an outdoor thermographic inspection in

Fig. 10. Intrinsic uncertainty test bench, using a blackbody source Fluke 4181.
4.2.3. General assumptions
In this work, the angular dimension of all analyzed objects is
larger than MSR. Thus the influence of MSR on measurement
uncertainty was neglected. TS was considered negligible as well.
DTR was estimated by the temperature range of Flir SC660 and
all the other sources of intrinsic uncertainty were experimentally
estimated in laboratory using a blackbody source Fluke 4181 (see
Fig. 10) under controlled conditions.
5. Results
5.1. Laboratory results
electrical connectors of a substation to exemplify the drawbacks
of using comparative methods for decision making.
Two electrical disconnectors (A and B) are shown in Fig. 12. At
the ends is possible to identify conductor cables and electrical con-
nectors that link the equipment to the system. Disconnector A is
being evaluated, whereas disconnector B is the reference. We mea-
sured temperature in two points for each component, T p1 and T p2
for disconnector A, T ref 1 and T ref 2 for disconnector B. The subscripts
1 and 2 mean cable and connector, respectively.
In a first analysis, we considered an emissivity of 0.75, which is
an arbitrary value usually chosen for inspections in Brazil. After, we
considered the emissivity of rusty aluminum e ¼ 0:30, obtained
from the literature [17], to enrich the discussion.
Table 5 shows the temperature measurements T p ; T ref and DT ref
evaluated by GUM and MCM, considering the emissivity of 0.75.
DT ref 2 value is 15:2 K with confidence level from 13:2 K to 17:3 K.
Using NETA standard (see Table 1) as reference DT ref 2 might be
considered, simultaneously, as ‘‘repair as time permits” (4 "C–
15 "C), and as ‘‘replace immediately” (>15 "C), leading to an
ambiguous conclusion.
If emissivity is 0.30, as shown in Table 6, temperature rises 11%
and 8.2% in terms of T p1 and T p2 , respectively, and uncertainty

Fig. 11 shows the temperature for both connectors, C1 and C2,

in function of current considering two different conditions, regular
(Fig. 11a) and coated (Fig. 11b) surfaces. Vertical bars denote
uncertainty ranges for a 95% of confidence level. As shown in
Fig. 11a, connectors C1 and C2 present uncertainties close to
&80 K and &70 K for a 450 A electrical current. This result contra-
dicts the affirmation that thermography accuracy improves in
inspections of deteriorated connectors. Fig. 11a also shows that
for 150 A, 200 A, 250 A, 300 A, 350 A and 400 A, the temperature
uncertainty intervals of connectors C1 and C2 partially overlap.
Under the metrological point of view, this particular result indi-
cates that is not possible to guarantee a temperature distinction
(neither deterioration stage) between both connectors at these
electrical current values, which is an incorrect conclusion since
C1 is highly degraded and C2 has not been used before.
On the other hand, coated regions at the connectors C1 and C2
present low uncertainty intervals, with approximately &9 K and Fig. 12. Thermography inspection in electrical substation.

Fig. 11. Temperature and electrical current relation for a) C1 (e ¼ 0:13) and C2 (e ¼ 0:12) b) C1 and C2 (e ¼ 0:97).
270 R.A.M. Ferreira et al. / Measurement 132 (2019) 263–271

Table 5
T p ; T ref and DT ref results for points 1 and 2 using e ¼ 0:75 (see Fig. 12).
Method T ob , K uðT ob Þ, K Confidence (95%) dlow dhigh Method T ob , K uðT ob Þ, K Confidence (95%) dlow dhigh

T p1 T p2

GUM 329.7 1.3 [327.3 332.2] 0.1 0.0 GUM 319.4 1.0 [317.5 321.2] 0.2 0.1
MCM 329.7 1.3 [327.4 332.2] MCM 319.4 0.9 [317.7 321.2]
T ref 1 T ref 2

GUM 301.2 0.5 [300.3 302.2] 0.1 0.1 GUM 304.2 0.5 [303.1 305.2] 0.1 0.1
MCM 301.2 0.6 [300.2 302.3] MCM 304.2 0.5 [303.2 305.1]
DT ref 1 DT ref 2

GUM 28.5 1.4 [25.9 31.2] 0.0 0.1 GUM 15.2 1.1 [13.1 17.4] 0.2 0.1
MCM 28.6 1.4 [25.9 31.3] MCM 15.2 1.1 [13.2 17.3]

Table 6
T p ; T ref and DT ref results for points 1 and 2 using e ¼ 0:30 (see Fig. 12).
Method T ob , K uðT ob Þ, K Confidence (95%) dlow dhigh Method T ob , K uðT ob Þ, K Confidence (95%) dlow dhigh

T p1 T p2

GUM 365.7 5.3 [355.4 376.2] 2.1 0.2 GUM 345.2 4.0 [337.4 353.3] 1.1 0.2
MCM 366.2 5.4 [357.5 376.0] MCM 345.5 4.1 [338.5 353.1]
T ref 1 T ref 2

GUM 305.3 1.6 [302.3 308.4] 0.6 0.8 GUM 312.3 1.9 [308.7 316.0] 0.2 0.7
MCM 305.4 1.9 [301.8 309.2] MCM 312.4 2.1 [308.5 316.6]
DT ref 1 DT ref 2

GUM 60.5 5.2 [49.6 71.3] 1.3 0.2 GUM 32.9 4.4 [24.3 41.4] 0.4 0.6
MCM 60.8 5.7 [50.9 71.4] MCM 33.1 4.6 [19.2 42.0]

Fig. 13. Source of uncertainty contributions (a) T p1 and (b) T p2 of disconnector A considering e ¼ 0:75.

Fig. 14. Source of uncertainty contributions (a) T p1 and (b) T p2 of disconnector A, considering e ¼ 0:30.
 R.A.M. Ferreira et al. / Measurement 132 (2019) 263–271
increases critically. In both cases (e ¼ 0:30 or 0.75), we recommend
to analyze the sources of uncertainty separately, looking to a better
diagnosis.
Fig. 13 shows the contributions of each uncertainty source for
T p1 and T p2 , considering e ¼ 0:75. In both T p1 and T p2 , the surface
emissivity is responsible for up to 40%, and atmospheric tempera-
ture and target distance are not representative. Fig. 14 shows the
same analysis for the emissivity of 0.30. It is possible to see that
for a lower emissivity values its contribution to uncertainty
becomes even more significant. Additionally, regarding the numer-
ical tolerance (see Table 6), GUM shows to be ineffective for the
emissivity of 0.30 (jdhigh # dlow j P d), providing a low quality confi-
dence level.
6. Conclusions
In this article we applied uncertainty analysis to the results of
thermographic inspections in electrical connectors during tests in
laboratory and outdoor conditions. Using GUM and MCM methods,
we demonstrate the high influence that emissivity has on temper-
ature uncertainty intervals, especially in low emissivity surfaces.
A test bench was mounted to simulate operational conditions of
electrical connectors and we performed thermographic inspections
in steady-state conditions for two H-type connectors in very differ-
ent degradation levels. We could conclude that the uncertainties
intervals partially overlap at a wide range of current values in
the non coated surfaces during the tests. This result contradicts
the affirmation that thermography accuracy improves in inspec-
tions of deteriorated connectors.
As expected, emissivity impacts the temperature uncertainty
highly, reinforcing the fact that adopting emissivity values without
any scientific support is not a recommended practice. In case of
low emissivity surfaces, the outdoor inspection indicated that the
results accuracy was compromised due to the high uncertainty
involved, and comparative methods may lead to incorrect
diagnosis.
Uncertainty sources other than emissivity did not represent a
great influence on the results accuracy, in special atmospheric
temperature and target distance, considering the experiments per-
formed. Regarding the method to determine uncertainty, Monte
Carlo Method (MCM) shows to be more appropriate than GUM,
mainly for low values of emissivity. In this case, GUM does not
meet the criteria established by the Joint Committee for Guides
in Metrology.
Acknowledgement
This project was supported by P&D 426 ANEEL (Agência Nacio-
nal de Energia Elétrica). The authors gratefully acknowledge the
financial support received from CNPq and CAPES.
References
[1] ANSI/NETA MTS, Standard for Maintenance Testing Specifications for Electrical
Power Equipment and Systems, International Electrical Testing Association,
2011.
271
[2] A. Ferrero, M. Prioli, S. Salicone, The construction of joint possibility
distributions of random contributions to uncertainty, IEEE Trans. Instrum.
Meas. 63 (1) (2014) 80–88.
[3] A. Taheri-Garavand, H. Ahmadi, M. Omid, S.S. Mohtasebi, K. Mollazade, A.J.R.
Smith, G.M. Carlomagno, An intelligent approach for cooling radiator fault
diagnosis based on infrared thermal image processing technique, Appl. Therm.
Eng. 87 (2015) 434–443.
[4] R. Morello, Potentialities and limitations of thermography to assess landslide
risk, Measurement 116 (2018) 658–668.
[5] F. Wang, Y. Wang, J. Liu, Y. Wang, Theoretical and experimental study on
carbon/epoxy facings-aluminum honeycomb sandwich structure using lock-in
thermography, Measurement 126 (2018) 110–119.
[6] C.M. Basheer, C.V. Krishnamurthy, K. Balasubramaniam, Hot-rod
thermography for in-plane thermal diffusivity measurement, Measurement
103 (2017) 235–240.
[7] C. Capua, R. Morello, I. Jablonski, Active and eddy current pulsed thermography
to detect surface crack and defect in historical and archaeological discoveries,
Measurement 116 (2018) 676–684.
[8] P.R. Muniz, R.A. Kalid, S.P.N. Cani, R.S. Magalhães, Handy method to estimate
uncertainty of temperature measurement by infrared thermography, Opt. Eng.
53 (7) (2014) 074101.
[9] W. Minkina, S. Dudzik, Simulation analysis of uncertainty of infrared camera
measurement and processing path, Measurement 39 (8) (2006) 758–763.
[10] K. Chrzanowski, R. Matyszkiel, J. Fischer, J. Barela, Uncertainty of temperature
measurement with thermal cameras, Opt. Eng. 40 (6) (2001) 1106–1114.
[11] M. Musto, G. Rotondo, M. Cesare, A. Del Vecchio, L. Savino, F. Filippis, Error
analysis on measurement temperature by means dual-color thermography
technique, Measurement 90 (2016) 265–277.
[12] M.S. Jadin, S. Taib, Recent progress in diagnosing the reliability of electrical
equipment by using infrared thermography, Infrared Phys. Technol. 55 (4)
(2012) 236–245.
[13] A.S.N. Huda, S. Taib, Application of infrared thermography for predictive/
preventive maintenance of thermal defect in electrical equipment, Appl.
Therm. Eng. 61 (2) (2013) 220–227.
[14] M.A. Mendes, L.G.R. Tonini, P.R. Muniz, C.B. Donadel, Thermographic analysis
of parallelly cables: a method to avoid misdiagnosis, Appl. Therm. Eng. 104
(2016) 231–236.
[15] A.S.N. Huda, S. Taib, Suitable features selection for monitoring thermal
condition of electrical equipment using infrared thermography, Infrared
Phys. Technol. 61 (2013) 184–191.
[16] B.P.A. Silva, R.A.M. Ferreira, S.C. Gomes Jr., F.A.R. Calado, R.M. Andrade, M.P.
Porto, On-rail solution for autonomous inspections in electrical substations,
Infrared Phys. Technol. 90 (2018) 53–58.
[17] FLIR Systems, User’s manual FLIR B6XX series FLIR P6XX series FLIR SC6XX
series, Number 1558550 in 1st ed., FLIR, 2010.
[18] Joint Committee for Guides in Metrology, JCGM 200: 2012 International
Vocabulary of Metrology Basic and General Concepts and Associated Terms,
third ed., JCGM, 2012.
[19] W. Minkina, S. Dudzik, Infrared Thermography: Errors and Uncertainties, first
ed., Wiley, 2009.
[20] P.M. Harris, M.G. Cox, On a monte carlo method for measurement uncertainty
evaluation and its implementation, Metrologia 51 (4) (2014) S176–S182.
[21] C. Wang, Z. Qiu, Y. Yang, Uncertainty propagation of heat conduction problem
with multiple random inputs, Int. J. Heat Mass Transf. 99 (2016) 95–101.
[22] Joint Committee for Guides in Metrology, Evaluation of Measurement Data
Supplement 1 to The Guide to the Expression of Uncertainty in Measurement –
Propagation of Distributions Using a Monte Carlo Method, first ed., JCGM,
2008.
[23] S. Tezuka, Uniform Random Numbers – Theory and Practice, Springer, US,
1995.
[24] K. Chrzanowski, Evaluation of thermal cameras in quality systems according to
ISO 9000 or EN45000 standards, Proc. SPIE 4360 (2001) 387–401.
[25] International Organization of Legal Metrology, OIML R141: Procedure for
Calibration and Verification of the Main Characteristics of Thermographic
Instruments, OIML, 2008.
[26] B. Lehmann, K.G. Wakili, Th. Frank, B.V. Collado, Ch. Tanner, Effects of
individual climatic parameters on the infrared thermography of buildings,
Appl. Energy 110 (2013) 29–43.
[27] A. Kylili, P.A. Fokaides, P. Christou, S.A. Kalogirou, Infrared thermography (IRT)
applications for building diagnostics: a review, Appl. Energy 134 (2014) 531–
549.